Low volatility and minimum variance are different portfolios

One sorts securities; the other combines them

A low-volatility screen ranks individual stocks by estimated variability. A minimum-variance portfolio chooses weights using both each stock's variance and its covariance with other holdings. A higher-volatility stock can reduce total portfolio risk if it diversifies the rest. Treating the two methods as synonyms hides the most important input to portfolio risk: how returns move together.

MSCI describes its minimum-volatility indexes as constrained volatility-minimizing constructions. That supports the distinction between a risk objective and a simple stock ranking; it does not validate a particular home-built covariance model. Any independent study must specify its own estimation window, return frequency, constraints and rebalancing schedule. An optimizer's output is conditional on those choices, not a forecast with guaranteed precision.

A two-stock calculation makes the difference visible

Let fictional Stock F have annualized volatility of 10% and Stock G have 20%, with zero correlation. A 50/50 portfolio has variance of 0.5 squared times 0.10 squared plus 0.5 squared times 0.20 squared, or 0.0125. Its volatility is about 11.18%, higher than holding F alone. Equal weighting is not the minimum-variance solution here.

For two uncorrelated assets without additional constraints, inverse-variance weights allocate 80% to F and 20% to G. Portfolio variance becomes 0.008 and volatility about 8.94%, below F's 10%. Change the assumed correlation to positive one and diversification disappears: the same 80/20 weights imply 12% volatility. The attractive result depends on a covariance assumption, not just a spreadsheet formula.

Estimation and constraints belong in the method

Historical covariance estimates are noisy, especially when the number of securities is large relative to observations. A model may regularize those estimates or use a factor structure, but its parameters must be fit using the available training history. Keep a simple weighting baseline. If small input changes cause large position changes, the supposed precision is creating turnover and implementation risk.

Constraints define what the optimizer is allowed to do. Long-only weights, issuer caps, sector limits and turnover budgets can prevent economically implausible allocations, while also moving the solution away from the unconstrained minimum. Record which constraints bind and how infeasibility is handled. Silently relaxing a cap until the solver succeeds makes historical portfolios incomparable and undermines any promised risk limit.

Lower estimated variance does not mean protected capital

Volatility measures dispersion, not the probability of meeting an investor's liability or avoiding a particular drawdown. A concentrated set of historically quiet stocks can still respond together to an earnings or interest-rate shock. A lower-beta portfolio can also lag a rising market. Evaluate realized concentration, drawdowns and changing correlations rather than describing low volatility as a substitute for loss protection.

Validate the full sequence: estimate covariance, solve weights, trade under declared assumptions, and measure the next period without refitting on its outcome. Compare predicted and realized risk by period, not only over the full sample. The figures here are mathematical illustrations with fixed inputs. They establish neither an empirical low-risk premium nor the future performance of an index, fund or strategy.

Inspect the example

Sources & reproduction

Primary references support the definitions and method. Numerical examples are illustrative unless a source record is explicitly identified.

• MSCI: Minimum Volatility Indexes

Continue the review